primal-barracuda 0.4.0

barraCuda — BARrier-free Rust Abstracted Cross-platform Unified Dimensional Algebra. Vendor-agnostic GPU compute via WGSL.
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
// SPDX-License-Identifier: AGPL-3.0-or-later
//! NAK/NVK stress tests — patterns from hotSpring Yukawa handoff.
//! Extracted from `tests.rs` for maintainability (wateringHole 1000-line convention).

use super::*;

/// Yukawa-pattern: compound assignments (+=, -=) on f64 accumulators.
/// hotSpring found that naga-guided rewrite of compound assignments
/// produced invalid SPIR-V on NVK/NAK. Verify our rewriter handles them.
#[test]
fn test_nak_compound_assignment() {
    let wgsl = r"
@group(0) @binding(0) var<storage, read> input: array<f64>;
@group(0) @binding(1) var<storage, read_write> output: array<f64>;
@group(0) @binding(2) var<uniform> n: u32;

@compute @workgroup_size(256)
fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
let i = gid.x;
if i >= n { return; }
var acc: f64 = 0.0;
for (var j = 0u; j < n; j = j + 1u) {
    if i == j { continue; }
    let dx = input[j] - input[i];
    let r2 = dx * dx;
    acc += dx / r2;
}
output[i] = acc;
}
";
    let result = rewrite_f64_infix_full(wgsl);
    assert!(
        result.is_ok(),
        "Yukawa compound-assign pattern should rewrite: {result:?}"
    );
    let src = result.unwrap();
    assert!(
        src.contains("_df64_sub_f64(") || src.contains("_df64_add_f64("),
        "should contain bridge calls, got:\n{src}"
    );
}

/// Yukawa-pattern: f64 comparisons (>, <) with continue in loops.
#[test]
fn test_nak_comparison_with_continue() {
    let wgsl = r"
@group(0) @binding(0) var<storage, read> pos: array<f64>;
@group(0) @binding(1) var<storage, read_write> forces: array<f64>;
@group(0) @binding(2) var<uniform> n: u32;

@compute @workgroup_size(256)
fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
let i = gid.x;
if i >= n { return; }
var fx: f64 = 0.0;
let cutoff_sq: f64 = 9.0;
for (var j = 0u; j < n; j = j + 1u) {
    if i == j { continue; }
    let dx = pos[j] - pos[i];
    let r2 = dx * dx;
    if r2 > cutoff_sq { continue; }
    let r = sqrt(r2);
    let force = exp(-r) / r2;
    fx += force * dx;
}
forces[i] = fx;
}
";
    let result = rewrite_f64_infix_full(wgsl);
    assert!(
        result.is_ok(),
        "comparison+continue pattern should rewrite: {result:?}"
    );
    let src = result.unwrap();
    assert!(
        src.contains("_df64_gt_f64(") || src.contains("_df64_mul_f64("),
        "should contain comparison or arithmetic bridges, got:\n{src}"
    );
}

/// Full Yukawa-like force kernel with all NAK-problematic patterns:
/// compound +=/-=, f64 comparisons, continue, sqrt, exp, nested ops.
/// Must produce valid WGSL when combined with `df64_core` + `df64_transcendentals`.
#[test]
fn test_nak_yukawa_full_validates() {
    let wgsl = r"
struct Params {
n: u32,
_pad0: u32,
cutoff: f64,
cutoff_sq: f64,
}

@group(0) @binding(0) var<storage, read> positions: array<f64>;
@group(0) @binding(1) var<storage, read_write> forces: array<f64>;
@group(0) @binding(2) var<storage, read_write> pe_buf: array<f64>;
@group(0) @binding(3) var<uniform> params: Params;

@compute @workgroup_size(256)
fn yukawa_force(@builtin(global_invocation_id) gid: vec3<u32>) {
let i = gid.x;
if i >= params.n { return; }

let xi = positions[i * 3u];
let yi = positions[i * 3u + 1u];
let zi = positions[i * 3u + 2u];

var fx: f64 = 0.0;
var fy: f64 = 0.0;
var fz: f64 = 0.0;
var pe: f64 = 0.0;

for (var j = 0u; j < params.n; j = j + 1u) {
    if i == j { continue; }

    let dx = positions[j * 3u] - xi;
    let dy = positions[j * 3u + 1u] - yi;
    let dz = positions[j * 3u + 2u] - zi;
    let r2 = dx * dx + dy * dy + dz * dz;

    if r2 > params.cutoff_sq { continue; }

    let r = sqrt(r2);
    let exp_r = exp(-r);
    let inv_r = 1.0 / r;
    let force_mag = exp_r * inv_r * inv_r * (1.0 + r);

    fx += force_mag * dx * inv_r;
    fy += force_mag * dy * inv_r;
    fz += force_mag * dz * inv_r;
    pe += exp_r * inv_r;
}

forces[i * 3u] = fx;
forces[i * 3u + 1u] = fy;
forces[i * 3u + 2u] = fz;
pe_buf[i] = pe * 0.5;
}
";
    let rewritten = rewrite_f64_infix_full(wgsl);
    assert!(
        rewritten.is_ok(),
        "Yukawa force kernel should rewrite successfully: {rewritten:?}"
    );

    let src = rewritten.unwrap();
    assert!(src.contains("_df64_mul_f64("), "should have mul bridges");
    assert!(src.contains("_df64_add_f64("), "should have add bridges");
    assert!(src.contains("_df64_sub_f64("), "should have sub bridges");
    assert!(src.contains("_df64_div_f64("), "should have div bridges");
    assert!(
        src.contains("_df64_gt_f64("),
        "should have comparison bridges"
    );

    // Validate the rewritten source compiles with df64 core
    const DF64_CORE: &str = include_str!("../../../shaders/math/df64_core.wgsl");
    const DF64_TRANS: &str = include_str!("../../../shaders/math/df64_transcendentals.wgsl");
    let full = format!("{DF64_CORE}\n{DF64_TRANS}\n{src}");

    let module = naga::front::wgsl::parse_str(&full)
        .unwrap_or_else(|e| panic!("should parse: {e}\n\nSource:\n{full}"));
    let mut v = naga::valid::Validator::new(
        naga::valid::ValidationFlags::all(),
        naga::valid::Capabilities::all(),
    );
    v.validate(&module)
        .unwrap_or_else(|e| panic!("should validate: {e}\n\nSource:\n{full}"));
}

#[test]
fn test_nak_cg_solver_validates() {
    let wgsl = r"
struct CgParams {
n: u32,
_pad0: u32,
_pad1: u32,
_pad2: u32,
}

@group(0) @binding(0) var<storage, read> r: array<f64>;
@group(0) @binding(1) var<storage, read> p: array<f64>;
@group(0) @binding(2) var<storage, read_write> x: array<f64>;
@group(0) @binding(3) var<uniform> params: CgParams;

@compute @workgroup_size(256)
fn cg_update_xr(@builtin(global_invocation_id) gid: vec3<u32>) {
let i = gid.x;
if i >= params.n { return; }

let alpha: f64 = 0.5;
let beta: f64 = 0.3;

x[i] = x[i] + alpha * p[i];

let r_new = r[i] - alpha * p[i];
let p_new = r_new + beta * p[i];
let dot = r_new * r_new;
let norm = sqrt(dot);

x[i] = x[i] + norm * 0.001;
}
";
    let rewritten = rewrite_f64_infix_full(wgsl);
    assert!(
        rewritten.is_ok(),
        "CG solver kernel should rewrite successfully: {rewritten:?}"
    );

    let src = rewritten.unwrap();
    assert!(src.contains("_df64_add_f64("), "should have add bridges");
    assert!(src.contains("_df64_mul_f64("), "should have mul bridges");
    assert!(src.contains("_df64_sub_f64("), "should have sub bridges");

    const DF64_CORE: &str = include_str!("../../../shaders/math/df64_core.wgsl");
    const DF64_TRANS: &str = include_str!("../../../shaders/math/df64_transcendentals.wgsl");
    let full = format!("{DF64_CORE}\n{DF64_TRANS}\n{src}");

    let module = naga::front::wgsl::parse_str(&full)
        .unwrap_or_else(|e| panic!("should parse: {e}\n\nSource:\n{full}"));
    let mut v = naga::valid::Validator::new(
        naga::valid::ValidationFlags::all(),
        naga::valid::Capabilities::all(),
    );
    v.validate(&module)
        .unwrap_or_else(|e| panic!("should validate: {e}\n\nSource:\n{full}"));
}

#[test]
fn test_nak_yukawa_celllist_validates() {
    let wgsl = r"
struct CellParams {
n: u32,
n_cells: u32,
cutoff: f64,
cutoff_sq: f64,
box_x: f64,
box_y: f64,
box_z: f64,
}

@group(0) @binding(0) var<storage, read> positions: array<f64>;
@group(0) @binding(1) var<storage, read> cell_list: array<u32>;
@group(0) @binding(2) var<storage, read> cell_start: array<u32>;
@group(0) @binding(3) var<storage, read_write> forces: array<f64>;
@group(0) @binding(4) var<uniform> params: CellParams;

fn pbc_wrap(dx: f64, box_len: f64) -> f64 {
var d = dx;
if d > box_len * 0.5 { d -= box_len; }
if d < -box_len * 0.5 { d += box_len; }
return d;
}

@compute @workgroup_size(256)
fn yukawa_celllist(@builtin(global_invocation_id) gid: vec3<u32>) {
let i = gid.x;
if i >= params.n { return; }

let xi = positions[i * 3u];
let yi = positions[i * 3u + 1u];
let zi = positions[i * 3u + 2u];

var fx: f64 = 0.0;
var fy: f64 = 0.0;
var fz: f64 = 0.0;

for (var c = 0u; c < params.n_cells; c++) {
    let start = cell_start[c];
    let end = cell_start[c + 1u];

    for (var idx = start; idx < end; idx++) {
        let j = cell_list[idx];
        if i == j { continue; }

        var dx = pbc_wrap(positions[j * 3u] - xi, params.box_x);
        var dy = pbc_wrap(positions[j * 3u + 1u] - yi, params.box_y);
        var dz = pbc_wrap(positions[j * 3u + 2u] - zi, params.box_z);
        let r2 = dx * dx + dy * dy + dz * dz;

        if r2 > params.cutoff_sq { continue; }

        let r = sqrt(r2);
        let exp_r = exp(-r);
        let inv_r = 1.0 / r;
        let f_mag = exp_r * inv_r * inv_r * (1.0 + r);

        fx += f_mag * dx * inv_r;
        fy += f_mag * dy * inv_r;
        fz += f_mag * dz * inv_r;
    }
}

forces[i * 3u] = fx;
forces[i * 3u + 1u] = fy;
forces[i * 3u + 2u] = fz;
}
";
    let rewritten = rewrite_f64_infix_full(wgsl);
    assert!(
        rewritten.is_ok(),
        "Yukawa cell-list kernel should rewrite: {rewritten:?}"
    );

    let src = rewritten.unwrap();
    const DF64_CORE_CL: &str = include_str!("../../../shaders/math/df64_core.wgsl");
    const DF64_TRANS_CL: &str = include_str!("../../../shaders/math/df64_transcendentals.wgsl");
    let full = format!("{DF64_CORE_CL}\n{DF64_TRANS_CL}\n{src}");

    let module = naga::front::wgsl::parse_str(&full)
        .unwrap_or_else(|e| panic!("should parse: {e}\n\nSource:\n{full}"));
    let mut v = naga::valid::Validator::new(
        naga::valid::ValidationFlags::all(),
        naga::valid::Capabilities::all(),
    );
    v.validate(&module)
        .unwrap_or_else(|e| panic!("should validate: {e}\n\nSource:\n{full}"));
}

/// Validate the **production** `yukawa_df64.wgsl` shader parses and validates
/// when combined with the DF64 preamble. This shader uses `df64_*` functions
/// directly (pre-rewritten) and is the actual code dispatched on NVK hardware.
#[test]
fn test_production_yukawa_df64_validates() {
    const DF64_CORE: &str = include_str!("../../../shaders/math/df64_core.wgsl");
    const DF64_TRANS: &str = include_str!("../../../shaders/math/df64_transcendentals.wgsl");
    const YUKAWA_DF64: &str = include_str!("../../../ops/md/forces/yukawa_df64.wgsl");

    // The production shader needs `round_f64` from the math_f64 preamble.
    // Inject a minimal stub — the real preamble is injected by `compile_shader_f64`.
    let round_stub = "fn round_f64(x: f64) -> f64 { return round(x); }\n";

    let full = format!("{DF64_CORE}\n{DF64_TRANS}\n{round_stub}\n{YUKAWA_DF64}");

    let module = naga::front::wgsl::parse_str(&full)
        .unwrap_or_else(|e| panic!("production yukawa_df64.wgsl should parse: {e}"));
    let mut v = naga::valid::Validator::new(
        naga::valid::ValidationFlags::all(),
        naga::valid::Capabilities::all(),
    );
    v.validate(&module)
        .unwrap_or_else(|e| panic!("production yukawa_df64.wgsl should validate: {e}"));
}

/// CPU reference implementation for Yukawa DF64 force/PE verification.
/// Computes the same quantities as `yukawa_df64.wgsl` using exact f64 arithmetic.
/// Use this to validate GPU results when NVK hardware is available.
#[test]
fn test_yukawa_cpu_reference_two_particles() {
    let kappa = 2.0_f64;
    let prefactor = 1.0_f64;
    let box_side = 10.0_f64;

    // Two particles along x-axis separated by distance 1.0
    let positions = [0.0, 0.0, 0.0, 1.0, 0.0, 0.0];
    let n = 2usize;

    let mut forces = [0.0f64; 6];
    let mut pe = [0.0f64; 2];

    for i in 0..n {
        let xi = positions[i * 3];
        let yi = positions[i * 3 + 1];
        let zi = positions[i * 3 + 2];

        let mut fx = 0.0f64;
        let mut fy = 0.0f64;
        let mut fz = 0.0f64;
        let mut pe_i = 0.0f64;

        for j in 0..n {
            if i == j {
                continue;
            }
            let dx = positions[j * 3] - xi;
            let dy = positions[j * 3 + 1] - yi;
            let dz = positions[j * 3 + 2] - zi;

            // PBC minimum image
            let dx = box_side.mul_add(-(dx / box_side).round(), dx);
            let dy = box_side.mul_add(-(dy / box_side).round(), dy);
            let dz = box_side.mul_add(-(dz / box_side).round(), dz);

            let r_sq = dx * dx + dy * dy + dz * dz;
            let r = r_sq.sqrt();
            let screening = (-kappa * r).exp();
            let kappa_r = kappa * r;
            let force_mag = prefactor * screening * (1.0 + kappa_r) / r_sq;
            let inv_r = 1.0 / r;

            // Force: repulsive (subtract, same sign as shader)
            fx = (force_mag * dx).mul_add(-inv_r, fx);
            fy = (force_mag * dy).mul_add(-inv_r, fy);
            fz = (force_mag * dz).mul_add(-inv_r, fz);

            // PE: U = prefactor * exp(-kappa*r) / r * 0.5 (half-counting)
            pe_i = (0.5 * prefactor * screening).mul_add(inv_r, pe_i);
        }
        forces[i * 3] = fx;
        forces[i * 3 + 1] = fy;
        forces[i * 3 + 2] = fz;
        pe[i] = pe_i;
    }

    // Analytical check: r=1, kappa=2, prefactor=1
    // |F| = prefactor * exp(-kappa*r) * (1 + kappa*r) / r^2 = exp(-2) * 3 ≈ 0.4060
    // Force is repulsive: particle 0 pushed away from particle 1 (negative x)
    let expected_force_mag = (-2.0f64).exp() * 3.0;
    assert!(
        (forces[0] + expected_force_mag).abs() < 1e-14,
        "particle 0 fx should be -mag (repulsive): {} vs expected {}",
        forces[0],
        -expected_force_mag
    );
    assert!(
        (forces[3] - expected_force_mag).abs() < 1e-14,
        "particle 1 fx should be +mag (repulsive opposite): {} vs {}",
        forces[3],
        expected_force_mag
    );

    // PE: 0.5 * exp(-2) / 1 = 0.5 * exp(-2) ≈ 0.0677
    let expected_pe = 0.5 * (-2.0f64).exp();
    assert!(
        (pe[0] - expected_pe).abs() < 1e-14,
        "pe[0]: {} vs expected {}",
        pe[0],
        expected_pe
    );
    assert!((pe[0] - pe[1]).abs() < 1e-14, "pe should be symmetric");

    // Newton's 3rd law
    assert!((forces[0] + forces[3]).abs() < 1e-14, "Newton 3rd law x");
    assert!((forces[1] + forces[4]).abs() < 1e-14, "Newton 3rd law y");
    assert!((forces[2] + forces[5]).abs() < 1e-14, "Newton 3rd law z");
}